CN104766333A - Vehicle door point welding robot path correction method based on stereoscopic vision - Google Patents

Abstract

The invention discloses a vehicle door point welding robot path correction method based on stereoscopic vision, and precise welding can be conducted by a welding robot intelligently at a welding point position. The method includes the following steps that 1 cameras are calibrated;after a vehicle door welding compensation dosage acquiring system based on the stereoscopic vision is established, before a vehicle door to be measured is sampled practically, the two cameras are calibrated by using a Zhengyou Zhang calibration method, and internal and external parameters of the cameras are acquired; 2 compensation dosage of the welding robot is acquired: practical edge folding lines of pressing plate pieces are imitated through points formed in edge folding lines, and the difference value between the vehicle door practical edge folding lines and theoretical edge folding lines is equal to the difference value between vehicle door practical welding point lines and theoretical welding point folding lines; 3 welding robot compensation dosage is imitated: after the welding robot compensation dosage is acquired, practical variable quantity of a stepping motor of the welding robot is obtained by using a kinematics inverse theory.

Description

Based on the car door spot welding robot course corrections method of stereoscopic vision

Technical field

The present invention relates to a kind of method for car door spot welding robot course corrections, more particularly, the present invention relates to a kind of car door spot welding robot course corrections method based on stereoscopic vision.

Background technology

Resistance spot welding in welding production process concentration of energy, sheet deformation is little, production efficiency is high, therefore be specially adapted to weld thin-walled parts, be one of the most frequently used welding method realizing automatic mass production at present, be therefore widely used in the welding field of automobile and auto parts and components.In modern cars manufacturing process; for auto manufacturing's product competition intensification, scale, mass, low cost, extensive, high-precision present situation; more and more higher requirement is proposed to being permitted the index such as multimember welding precision and speed; thus impelled promoting the use of of welding robot, the development and apply of Robotics is the important symbol of a national high-tech level and industrial production automation degree.The particularly fast development of Domestic Automotive Industry in recent years, impels resistance spot welding machine people constantly to promote at automotive field, use and develop.Technology of Welding Robot based on stereoscopic vision is a kind of new technology combining the multidisciplinary formation such as computing machine, kybernetics, information and sensing technology, is the developing direction of automobile industry welding field.

Car door parts are in welding process, the jig of welding work pieces is due to long effect of stress, the change in displacement of welding work pieces in motion process, can make the actual welding position of solder joint and theoretical welding position that deviation occurs, weld error when data shows and can reach 5mm-10mm, thus cause final solder joint to occur various defect, the strength and stiffness of welding are made not reach requirement, affect the Rigidity and strength of white car door, may be little on the normal traveling impact of automobile, but but leave potential safety hazard when automobile generation emergency condition; The quality of point quality not only affects the safety and reliability of car, also will affect presentation quality and the assembly precision of the white car door of car, and then affect quality and the usability of vehicle complete vehicle.In sum, the accurate location of white car door butt welding point position is before welding very urgent.

Traditional automobile spot welding robot, the path of welding of welding gun moves under program control, path is changeless, even if car door due to a variety of causes change time, the welding position of welding gun can not correct timely and effectively, thus after finally having welded, make actual welding position deviate from desirable welding position, thus cause various white car door point quality problem.Up to the present, the domestic scientific research institution that has or universities and colleges, the Cui Xingqiang of such as Hunan University, follows the tracks of for the vision of stitching weldering in welding and has had certain research, but due to the position singularity of weld nugget, the domestic research of the course corrections for spot welding at present does not almost have.

Summary of the invention

The invention provides a kind of car door spot welding robot course corrections method based on stereoscopic vision, make welding robot carry out intelligentized accurate welding in bond pad locations.

The object of the invention is to adopt following technical scheme to realize:

A kind of car door spot welding robot course corrections method based on stereoscopic vision, it is characterized in that, the method uses a kind of automobile door welding compensation rate based on stereoscopic vision to obtain system, comprise be arranged on door skin streamline track oblique upper a word laser instrument, be symmetricly set on word laser instrument both sides and be positioned at two video cameras directly over door skin streamline track, and the terminal device be connected with a word laser instrument and two video cameras;

This car door spot welding robot course corrections method comprises the following steps:

Step one, camera calibration: put up described based on after the automobile door welding compensation rate acquisition system of stereoscopic vision, before actual acquisition car door surface image to be measured, first Zhang Zhengyou scaling method is utilized to demarcate two video cameras, obtain the inside and outside parameter of video camera, to set up the nonlinear relationship between actual determinand locus and image coordinate, and then calculate the volume coordinate of detected image Feature point correspondence;

Step 2, welding robot compensation rate obtain: a word laser instrument Emission Lasers tilts to beat on the actual flange line of punching press plate, laser rays generation deviation, and break, just on flange line, simulates the actual flange line of punching press plate by the point on flange line; Namely difference between car door actual flange line and theoretical flange line is equivalent to the difference between car door actual solder joint line and theoretical solder joint line;

Step 3, simulation welding robot compensation rate realize: after acquisition welding robot compensation rate, utilize the inverse solution of robot kinematics theoretical, ask for the actual change amount of welding robot stepper motor, moved by programming Control stepper motor, finally realize welding robot and weld in correct position.

A kind of car door spot welding robot course corrections method based on stereoscopic vision of the present invention, wherein, the detailed process of step one camera calibration is:

The method utilizing Zhang Zhengyou to demarcate is demarcated, and adopt CAD Software on Drawing to demarcate masterplate, calibrating template is the gridiron pattern of 9 × 9, and its specification is 270mm × 270mm.And then two video cameras are fixed on appropriate location and keep motionless, constantly angle between masterplate and imaging plane and orientation are demarcated in conversion afterwards, and two video cameras gather 16 width images to demarcation masterplate.Feature corner extraction is carried out to camera calibration template image, each camera calibration template image can extract 100 angle points, 16 width images obtain 1600 characteristic angle dot image coordinates altogether, according to Zhang Zhengyou scaling method, world coordinate system is set on an angle point of the most left upper of cross-hatch pattern picture, Z-direction coordinate figure is taken as zero, because each demarcation masterplate size careful design is 30mm, therefore gridiron pattern feature angle point can accurately be learnt at the world coordinates of X and Y-direction, therefore after the volume coordinate obtaining plane picture characteristic angle point coordinate and corresponding point, tool box can be utilized to complete demarcation, calculate the internal and external parameter of video camera,

A kind of car door spot welding robot course corrections method based on stereoscopic vision of the present invention, wherein, step 2 welding robot compensation rate obtains and comprises following concrete steps:

1) Image semantic classification is carried out to car door punching press plate flange line image in camera review;

2) the actual flange line of punching press plate and the volume coordinate of laser rays intersection point is obtained;

3) rotate actual flange line and theoretical flange line, obtain welding compensation rate.

A kind of car door spot welding robot course corrections method based on stereoscopic vision of the present invention, wherein, step 1) Image semantic classification is carried out to car door punching press plate flange line image in camera review specifically comprise following process:

1.1) rgb image of camera acquisition is converted into gray level image;

Tilt to beat on the flange line of punching press plate by laser, adopt two camera acquisition images, obtain left and right two width images respectively, the original image gathered is RGB image, and RGB image is changed into gray level image, is changing into a data matrix;

1.2) obtain target area and image enhaucament is carried out to target area;

Adopt the cutting of image parameter formula to carry out shearing to target area to extract, obtain target area image; Image enhaucament is carried out to target area, carries out logical operation simultaneously, reduce integral image gray scale;

1.3) gradation of image adjustment;

Carry out the image after logical operation, laser rays region and surrounding environment contrast not obvious, therefore the adjustment of gray scale will be carried out to image, namely the unsharp image of script is become clear, or suppress some unwanted feature of image thus other target signature is enhanced, thus the visual effect of the image after process is enhanced;

1.4) image noise reduction;

In experimentation, due to experiment condition and the quantitative limitation of video camera matter, often make the image photographed be subject to the impact of picture noise, produce a little distortion, therefore in order to obtain the high result of precision comparison, medium filtering is adopted to carry out noise reduction process to the image after grey level enhancement;

1.5) binary conversion treatment of gray level image.

A kind of car door spot welding robot course corrections method based on stereoscopic vision of the present invention, wherein, step 2) volume coordinate of actual flange line and laser rays intersection point that obtains punching press plate specifically comprises following process:

2.1) to step 1) pretreated after image carry out Image erosion process, extract the skeleton of laser rays;

Image erosion is the most basic is also one of most important Morphological scale-space method, corrosion can melt the border of object, thus obtain the skeleton of required target, concrete corrosive effect is relevant with the shape of target image itself and structural element: if target area is greater than structural element on the whole, the result of corrosion makes edge, a target area disappearance part, and the size of disappearance subregion is determined by the size of structural element; If target area itself is less than structural element, then after carrying out erosion operation, target area will disappear completely.

2.2) Hough transformation obtains the image coordinate of the actual flange line of punching press plate and laser rays intersection point;

Hough (Hough) conversion is a kind of very important method detecting discontinuous point, and its principle marginal point is coupled together the method forming closed region.What adopt is the curvilinear equation that Hough (Hough) transfer pair skeleton carries out that matching obtains skeleton, asks for the actual flange line of punching press plate and laser rays intersection point image coordinate.

2.3) three-dimensional reconstruction is carried out to the actual flange line of pressed sheet part and laser rays intersection point;

Stereo matching can obtain the coordinate that two width image pixels are mutually corresponding about object, the coordinate of object under three-dimensional coordinate system is tried to achieve again according to the corresponding relation between two width images is reverse, be the three-dimensional reconstruction of spatial point, and then obtain the three dimensional space coordinate of intersection point.

A kind of car door spot welding robot course corrections method based on stereoscopic vision of the present invention, wherein, step 3) rotate actual flange line and theoretical flange line, obtain welding compensation rate and comprise following detailed process:

3.1) adopt the volume coordinate point of Matlab Curve Fitting Toolbox to the actual flange line of punching press plate obtained and laser rays intersection point to carry out matching, obtain a space curve, i.e. actual flange line;

Run Matlab software, obtained 3 d space coordinate value is inputted in Command Window order pane, click the Start order button in the lower left corner, Matlab interface, select Toolboxs tool box, Curve Fitting Tool (cftool) button in CurveTitting is found from drop-down menu, click working procedure Curve Fitting Toolbox, or direct command window directly inputs " cftool ", click carriage return, Curve Fitting Toolbox can be called, write x, y and z value of 3 d space coordinate respectively, the curve of matching can be obtained.

3.2) the theoretical flange line of punching press plate is obtained;

The flange line of the Catia model of car door extracts several points, and its volume coordinate is transformed in the middle of actual car door coordinate system, afterwards matching is carried out to these spatial point and obtain theoretical flange line;

3.3) rotate actual flange line and theoretical flange line, calculate welding compensation rate;

Actual flange line and theoretical flange line are rotated, projects to X-Y plane respectively, in X-Z plane and Z-Y plane, obtain the compensation rate of X-direction, Y-direction and Z-direction respectively.

Compared with prior art the invention has the beneficial effects as follows:

1. the car door spot welding robot course corrections method key point based on stereoscopic vision of the present invention is the size obtaining compensation rate, its prerequisite is the coordinate of the point of trying to achieve on actual flange line, have employed a kind of usage of novel line laser, tilt to beat on the actual flange line of punching press plate by laser, can deviation be there is in laser rays, break, just on actual flange line, can simulate actual flange line by the point on actual flange line.The method can determine the ignorant curvilinear equation in position easily and fast.

2. an innovative point of the car door spot welding robot course corrections method based on stereoscopic vision of the present invention is the calculating of compensation rate, is also that invention realizes judging that whether welding position is crucial accurately before welding.Car door can be similar to when carrying out door frame welding regards rigid body as, distance between sealing wire and flange line is fixing, when car door is moved, flange line and sealing wire are that entirety is moved, between distance constant, whether accurately so the displacement between solder joint line, namely compensation rate just changes into displacement variable between flange line, and then solve and judge welding position problem before welding.

3. the disposal route of car door spot welding robot course corrections method to structured light based on stereoscopic vision of the present invention is improved, pre-service processes image exactly, target area required for us is highlighted, when impact point is extracted, by the comparison to multiple method, the final selected Hough transform that adopts asks for impact point, ensures the accuracy of impact point.

Accompanying drawing explanation

Fig. 1 is the overview flow chart of the car door spot welding robot course corrections method based on stereoscopic vision of the present invention;

Fig. 2 is the schematic diagram that the car door spot welding robot course corrections system based on stereoscopic vision of the present invention obtains welding compensation rate online;

Fig. 3 is the calibrating template based on carrying out timing signal in the car door spot welding robot course corrections method of stereoscopic vision to video camera of the present invention;

Fig. 4 is the schematic diagram based on obtaining welding compensation rate in the car door spot welding robot course corrections method of stereoscopic vision of the present invention;

Fig. 5 is the particular flow sheet based on obtaining welding compensation rate in the car door spot welding robot course corrections method of stereoscopic vision of the present invention;

Fig. 6 be of the present invention based in the car door spot welding robot course corrections method of stereoscopic vision to the workflow diagram of medium filtering during image noise reduction in Image semantic classification;

Fig. 7-a is image space when adopting Hough transform in the volume coordinate based on obtaining the actual flange line of punching press plate and laser rays intersection point in the car door spot welding robot course corrections method of stereoscopic vision of the present invention;

Fig. 7-b is parameter space when adopting Hough transform in the volume coordinate based on obtaining the actual flange line of punching press plate and laser rays intersection point in the car door spot welding robot course corrections method of stereoscopic vision of the present invention;

Fig. 8 is the schematic diagram based on carrying out three-dimensional reconstruction in the car door spot welding robot course corrections method of stereoscopic vision to the actual flange line of pressed sheet part and laser rays intersection point of the present invention;

Fig. 9-a be of the present invention based in the car door spot welding robot course corrections method of stereoscopic vision to the schematic diagram at actual flange line and theoretical flange wire-wound X-axis rotation alpha angle;

Fig. 9-b is the schematic diagram based on rotating β angle in the car door spot welding robot course corrections method of stereoscopic vision to actual flange line and theoretical flange wire-wound Y-axis of the present invention;

Figure 10 is the screw rod schematic diagram adopted based on cross linear guides in the car door spot welding robot course corrections method of stereoscopic vision of the present invention;

Figure 11 is the schematic diagram based on asking for stepper motor rotational angle in the car door spot welding robot course corrections method of stereoscopic vision of the present invention;

In figure: 1. door skin streamline, 2. video camera, 3. an information processing terminal device, 4. a word generating laser, 5. No. two video cameras.

Embodiment

Below in conjunction with accompanying drawing, technical scheme of the present invention is explained in detail:

Consult Fig. 2, it is the rough schematic view of a kind of acquisition of the automobile door welding compensation rate based on stereoscopic vision system that the present invention designs in figure, when door skin 1 slowly moves along track, one word laser instrument 4 tilts to beat on the actual flange line of punching press plate laser, the video camera 2 utilizing word laser instrument 4 symmetria bilateralis to arrange and No. two video cameras 5 collect left and right two width images respectively, then are processed to information processing terminal device 3 by the image transmitting collected.

During detection, a video camera 2 and No. two video cameras 5 are all positioned at 1.5 meters of directly over car door, what a video camera 2 and No. two video cameras 5 adopted is the model that vision facilities company of Beijing Daheng produces is DH-HV1302UM-T, the CCTV & VIDEO camera lens of resolution to be 1248*1024 and focal length be 12.5-75mm; One word generating laser 4 is placed on car door diagonally forward 1.5 meters of, and use model to be sz-mt50i, output power is 40-50mW, and light luminance is high, relatively little by external environmental interference; One, computing machine.Consult Fig. 2, calibrating template is the gridiron pattern of 9 × 9, and its specification is 270mm × 270mm.

Consult Fig. 1, a kind of car door spot welding robot course corrections method based on stereoscopic vision, whole testing process is mainly divided into 3 stages: 1. the calibration phase of video camera; 2. welding robot compensation rate obtains the stage; 3. simulate the implementation phase of welding robot compensation rate.

1. the calibration phase of video camera:

After putting up detection platform (the automobile door welding compensation rate based on stereoscopic vision obtains system), in order to ensure carrying out smoothly of detection, before collection car door surface image to be measured, first must demarcate a video camera 2 and No. two video cameras 5, obtain the relevant inside and outside parameter of video camera, just can set up the nonlinear relationship of the correspondence between actual determinand locus and image coordinate, utilize this nonlinear relationship, the volume coordinate of detected image Feature point correspondence could be calculated.

Consult Fig. 3, method (the ZHANG Z Y utilizing Zhang Zhengyou to demarcate, A Flexible New Technique forCamera Calibration [R] .Microsoft Corporation, NSR-TR-98-71,1998) demarcate, first adopt CAD Software on Drawing to demarcate masterplate, calibrating template is the gridiron pattern of 9 × 9, and its specification is 270mm × 270mm.And then two video cameras 2,5 are fixed on appropriate location and keep motionless, the angle afterwards constantly between conversion demarcation masterplate and imaging plane and orientation, two video cameras gather 16 width images to demarcation masterplate.In the 16 width images obtain shooting, the video camera gridiron pattern target image in the lower left corner carries out Feature corner extraction, and each video camera gridiron pattern target image can extract 100 angle points, and 16 width images obtain 1600 characteristic angle dot image coordinates altogether.According to Zhang Zhengyou scaling method, world coordinate system is set on an angle point of the most left upper of video camera gridiron pattern target image, Z-direction coordinate figure is taken as zero, because each gridiron pattern size careful design is 30mm, therefore gridiron pattern feature angle point can accurately be learnt at the world coordinates of X and Y-direction.Therefore, after the volume coordinate obtaining gridiron pattern target image characteristic angle point coordinate and corresponding point in demarcation plane, tool box can be utilized to complete demarcation, calculate the internal and external parameter of video camera 2,5.

2. welding robot compensation rate obtains the stage:

Consult Fig. 4, after actual welding completes, have certain distance between solder joint and flange line, itself can be regarded as rigid body in car door local, the distance between solder joint and flange line is changeless.When car door is moved, the difference between actual solder joint line and theoretical solder joint line, just changes into the difference between actual flange line and theoretical flange line, obtains the difference between actual solder joint line and theoretical solder joint line, is exactly the compensation rate that welding robot needs.

1) Image semantic classification is carried out to the initial acquisition image of car door punching press plate flange line in camera review:

The image taken due to video camera 2,5 has more useless background, and image is easily subject to inner plate of car door surface reflection and noise effect, therefore need to reject useless background and reduce image processing time and complexity, and adjust image is reflective and carries out noise reduction process to noise, the step of the door skin surface initial detecting image of video camera 2,5 shooting being carried out to Image semantic classification is as follows:

1.1) rgb image of camera acquisition is converted into gray level image: tilt to beat on the flange line of punching press plate by laser, adopt two camera acquisition images, obtain left and right two width images respectively, the original image gathered is RGB image, i.e. rgb image, and RGB image is changed into gray level image, be changing into a data matrix, such meeting simplified operation process, improves counting yield, and then improves general image process operation efficiency.

1.2) obtain target area and image enhaucament is carried out to target area: shearing manipulation is carried out to the target area image in gray level image, obtain the local area image of car door flange line and laser rays in camera review, reject useless background area, image enhaucament is carried out to acquisition region, strengthen the contrast of target area inner laser line and peripheral region, carry out logical operation simultaneously, reduce integral image gray scale, improve arithmetic speed.

Under normal conditions, the scope of taking pictures of video camera is greater than required target zone, many unnecessary, unnecessary information can be included, and what the present invention adopted is that the cutting of image parameter formula carries out shearing extraction to target area, obtains and needs region;

After obtaining target area image, image enhaucament is carried out to target area, to strengthen the contrast of target area inner laser line and peripheral region, realize the further process to image and analysis, also improve the process of picture quality simultaneously, in order to improve arithmetic speed, logical operation being carried out to target image, reducing integral image gray scale.

1.3) gray scale adjustment is carried out to image, carry out the image after logical operation, laser rays region and surrounding environment contrast not obvious, therefore the adjustment of gray scale will be carried out to image, namely the unsharp image of script is become clear, or suppress some unwanted feature of image thus other target signature is enhanced, thus the visual effect of the image after process is enhanced.

1.4) image noise reduction;

In experimentation, due to experiment condition and the quantitative limitation of video camera matter, often make the image photographed be subject to the impact of picture noise, produce a little distortion, therefore in order to obtain the high result of precision comparison, the general medium filtering that adopts carries out noise reduction process to the image after grey level enhancement.

The effect of medium filtering extremely relies on the selection of filter window size, if window is too large, can make edge fog, the too little meeting of window makes noise reduction bad, and counting yield is lower.Therefore must improve the algorithm of medium filtering.

Consult Fig. 6, the principle of medium filtering is when filtering, during sequence, all pixel values are sorted each time and there is no need, for use such as formula (1) Suo Shi 3 × 3 template, when program from left to right moves in the drawings, each change be only tertial 3 values, then the values of two row have arranged in upper once operation, and therefore we only need in the sequence that this is orderly, utilize innovatory algorithm to inject tertial three successively and are newly worth.Known by the above analysis to ordering principle, only need to sort once to all grey scale pixel values in template window, then according to the moving direction of template, insert the 3rd in order and be worth.The present invention adopts bubble sort method, is once inserted in 6 ordered sequences by 3 grey scale pixel values, then adopts split to search insertion, at most only needs to compare for 10 times and can complete, and therefore adopt innovatory algorithm to improve counting yield, practicality is good.

$\left[\begin{array}{ccc}{k}_1& k_2& k_3\\ k_4& k_5& k_6\\ k_7& k_8& k_9\end{array}\right]---\left(1\right)$

1.5) binary conversion treatment of gray level image;

In order to improve the contrast of laser beam and surrounding environment, be convenient to rim detection, gray level image to be changed into bianry image, principle is from the pixel value gray level image, get a threshold value, be designated as Tc, the pixel being greater than this threshold value is attributed to a class, the pixel being less than this threshold value belongs to another kind of, the method is thresholding method, that is:

$f(x,y)=\left\{\begin{array}{cc}0& f(x,y)<T_c\\ 255& f(x,y)<T_c\end{array}\right.---\left(2\right)$

By the key that the known selection gray threshold Tc appropriately of binaryzation principle is image binaryzation process.If T _cselect improperly will cause image fault.Choosing method about threshold value generally has P parametric method, Two-peak method, histogram method.The present invention adopts histogram, and principle makes the showing with histogrammic formal intuition of the pixel value in gray level image, then carry out Threshold selection according to the statistical law of pixel grey scale.Histogram treatment is carried out to image, when there is two obvious peak values in grey level histogram, then in image, most grey scale pixel value concentrates near two peak values, and therefore can elect the pixel value near a peak value as target, the pixel near another peak value elects background as.The pixel value of low ebb between two peak values is only needed to be set to threshold value Tc, so just can be well separated by object and background.

2) volume coordinate of punching press plate flange line and laser rays intersection point is obtained

In order to obtain punching press plate flange line and laser rays intersection point information, need to carry out Morphological scale-space to pretreated image, the step of the volume coordinate of concrete acquisition punching press plate flange line and laser rays intersection point is as follows:

2.1) Image erosion process is carried out to the image after pretreated, extract the skeleton of laser rays;

Image erosion is the most basic is also one of most important Morphological scale-space method, corrosion can melt the border of object, thus obtain the skeleton of required target, concrete corrosive effect is relevant with the shape of target image itself and structural element: if target area is greater than structural element on the whole, the result of corrosion makes edge, a target area disappearance part, and the size of disappearance subregion is determined by the size of structural element; If target area itself is less than structural element, then after carrying out erosion operation, target area will disappear completely.

2.2) Hough transformation obtains the image coordinate of punching press plate flange line and laser rays intersection point;

Hough (Hough) conversion is a kind of very important method detecting discontinuous point, and its principle marginal point is coupled together the method forming closed region.What adopt is the curvilinear equation that Hough (Hough) transfer pair skeleton carries out that matching obtains skeleton, asks for punching press plate flange line and laser rays intersection point image coordinate.

The ultimate principle of Hough transform is the duality of dotted line.In image xy space, all straight lines crossing point (x, y) can write straight general expression y=kx+b, and wherein k is straight slope, and b is the intercept of straight line in y-axis.The general expression of straight line can be rewritten into b=-kx+y.Point (x is crossed in image space xy ₁, y ₁) straight-line equation be y ₁=kx ₁+ b, can be rewritten into b=-kx in parameter space kb ₁+ y ₁, expression be the straight line in parameter space kb.Cross point (x ₂, y ₂) straight-line equation be y ₂=kx ₂+ b, also can be write as b=-kx ₂+ y ₂, expression be the another straight line in parameter space kb.Suppose that these two straight lines intersect at a point in parameter space kb (k ', b '), a mistake (x here in point (k ', b ') correspondence image space xy ₁, y ₁) and (x ₂, y ₂) straight line, so this point meets y ₁=k ' x ₁+ b ' and y ₂=k ' x ₂+ b '.Point (x is crossed as can be seen here in image space xy ₁, y ₁) and (x ₂, y ₂) straight line on all corresponding parameter space kb of every bit in straight line, these straight line intersection are in point (k ', b ').

Consult Fig. 7, a () is image space, b () is parameter space, in image space the corresponding line intersected in parameter space of the point of conllinear, and all straight lines intersected at a point in parameter space have the point of conllinear corresponding with it in image space.The principle of Hough transform is exactly utilize the duality of this Points And lines that the test problems in image space is transformed among parameter space.

But in the middle of practical application, usually the polar form of the equation y=kx+b of straight line is replaced ρ=xcos θ+ysin θ.Wherein (ρ, θ) refer to one from initial point to line on the vector of closest approach, this vector is orthogonal relation with straight line.

Putting to find these straight line formed, ρ-θ space quantization can be divided into many little spaces, according to each (x ₀, y ₀) put the quantized value substituting into θ, calculate each ρ, when the value calculated drops on certain little space, so the counting totalizer in this little space adds 1, when after whole (x, y) point transformation, tested in little space, little spaces maximum for count value is corresponded to collinear point, and its (ρ, θ) makes the parameter of fitting a straight line.The usual corresponding non-colinear point in little space that count value is less, can give up need not.Consider the requirement of these two aspects, obtain most suitable quantized value.

The performing step of Hough transform is generally as follows:

A () carries out initial transformation to transform domain, suitable quantization parameter space

B (), by space how little for parameter space segmenting program, suppose that each little space is a totalizer, and its initial value is set to 0;

(c) for image space every bit (x, y), at it totalizer corresponding to the parametric equation that meets adds 1;

D () calculates spatial model parameter corresponding to maximal value in the middle of accumulator array.

After calculating completes, find the maximum point in the totalizer in ρ-θ transform domain, record the value of ρ and θ of each maximum point i.e. peak point, according to the ρ of the peak point of record, θ value repaints curve in rectangular coordinate system.

After straight-line segment marks, just require the intersection point of two line segments, intersection point is exactly corresponding punching press plate flange line and the intersection point of laser rays, forms system of equations by combining two straight-line equations, just can in the hope of the intersection point of two straight lines.If two straight line is respectively:

$\left\{\begin{array}{c}{\mathrm{\&rho;}}_1=x\cos \left({\mathrm{\&theta;}}_1\right)+y\sin \left({\mathrm{\&theta;}}_1\right)\\ {\mathrm{\&rho;}}_2=x\cos \left({\mathrm{\&theta;}}_2\right)+y\sin \left({\mathrm{\&theta;}}_2\right)\end{array}\right.---\left(3\right)$

Separate the intersection point that linear equation in two unknowns group can obtain two straight lines.

2.3) three-dimensional reconstruction is carried out to the intersection point of pressed sheet part flange line and laser rays;

General pattern, after process, can obtain a large amount of two-dimensional pixel coordinate informations, after Stereo matching, carry out three-dimensional reconstruction, and then obtains three dimensional space coordinate.Due to the negligible amounts of main points required for the present invention, and left and right cameras is taken pictures in a certain order, and the mutual corresponding point of each image mid point are easy to find.

Consult Fig. 8, according to the two-dimensional image information of correspondences more than two width or two width, obtaining the method for object dimensional geological information, is exactly the three-dimensional reconstruction based on stereoscopic vision.Object in space, take pictures through two cameras, namely two width projected images about object can be obtained, corresponding relation is had between information on this two width image, through Stereo matching, can obtain the coordinate that this two width image pixel is mutually corresponding, then try to achieve the coordinate of object under three-dimensional coordinate system according to the corresponding relation between two width images is reverse, the method is the three-dimensional reconstruction of spatial point.

An object is had, O in hypothesis space ₁, O ₂be the position at two video camera places, take pictures and obtain left and right two width image I ₁, I ₂, 1 P coordinate in three dimensions on object is [X, Y, Z] ^t, at left and right two width image I ₁, I ₂on subpoint be respectively P _land P _r, under image coordinate system, corresponding homogeneous coordinates are respectively [u ₁, v ₁, 1] ^t, [u ₂, v ₂, 1] ^t, then space mid point P and P _land P _rthere is following corresponding relation:

$Z_{c1}\left[\begin{array}{c}{u}_1\\ v_1\\ 1\end{array}\right]=\left[\begin{array}{cccc}{m}_{11}^1& m_{12}^1& m_{13}^1& m_{14}^1\\ m_{21}^1& m_{22}^1& m_{23}^1& m_{24}^1\\ m_{31}^1& m_{32}^1& m_{33}^1& m_{34}^1\end{array}\right]\left[\begin{array}{c}X\\ Y\\ Z\\ 1\end{array}\right]---\left(4\right)$

$Z_{c2}\left[\begin{array}{c}{u}_2\\ v_2\\ 1\end{array}\right]=\left[\begin{array}{cccc}{m}_{11}^2& m_{12}^2& m_{13}^2& m_{14}^2\\ m_{21}^2& m_{22}^2& m_{23}^2& m_{24}^2\\ m_{31}^2& m_{32}^2& m_{33}^2& m_{34}^2\end{array}\right]\left[\begin{array}{c}X\\ Y\\ Z\\ 1\end{array}\right]---\left(5\right)$

Wherein,

$M_L=\left[\begin{array}{cccc}{m}_{11}^1& m_{12}^1& m_{13}^1& m_{14}^1\\ m_{21}^1& m_{22}^1& m_{23}^1& m_{24}^1\\ m_{31}^1& m_{32}^1& m_{33}^1& m_{34}^1\end{array}\right]=A_l*\left[\begin{array}{cc}{R}_l& T_l\\ 0^T& 1\end{array}\right]---\left(6\right)$

$M_R=\left[\begin{array}{cccc}{m}_{11}^2& m_{12}^2& m_{13}^2& m_{14}^2\\ m_{21}^2& m_{22}^2& m_{23}^2& m_{24}^2\\ m_{31}^2& m_{32}^2& m_{33}^2& m_{34}^2\end{array}\right]=A_r*\left[\begin{array}{cc}{R}_r& T_r\\ 0^T& 1\end{array}\right]---\left(7\right)$

Wherein M _l, M _rbe respectively the outside matrix parameter of left and right cameras, A _l, A _rbe respectively the inner parameter of left images, $\left[\begin{array}{cc}{R}_l& T_l\\ 0^T& 1\end{array}\right],\left[\begin{array}{cc}{R}_r& T_r\\ 0^T& 1\end{array}\right]$ Be respectively the external parameter matrix of left and right cameras, R _l, R _rfor the rotation matrix of left and right cameras, T _l, T _rbe respectively the translation vector of left and right cameras.

Inner parameter and the external parameter of video camera are obtained by camera calibration hereinbefore, and the inside and outside parameter according to video camera can obtain projection matrix M _l, M _r, in expression formula 6 and 7, Z _c1and Z _c2unnecessary intermediate variable, therefore, cancellation Z _c1or Z _c2, can obtain about X, Y and Z tetra-systems of linear equations:

$\left\{\begin{array}{c}(u_1{m}_{31}^1-m_{11}^1)X+(u_1{m}_{32}^1-m_{12}^1)Y+(u_1{m}_{33}^1-m_{33}^1)Z=m_{14}^1-u_1{m}_{34}^1\\ (v_1{m}_{31}^1-m_{21}^1)X+(v_1{m}_{32}^1-m_{22}^1)Y+(v_1{m}_{33}^1-m_{23}^1)Z=m_{24}^1-v_1{m}_{34}^1\\ (u_2{m}_{31}^2-m_{11}^2)X+(u_2{m}_{32}^2-m_{12}^2)Y+(u_2{m}_{33}^2-m_{13}^2)Z=m_{14}^2-u_2{m}_{34}^2\\ (v_2{m}_{31}^2-m_{21}^2)X+(v_2{m}_{32}^2-m_{22}^2)Y+(v_2{m}_{33}^2-m_{23}^2)Z=m_{24}^2-v_2{m}_{34}^2\end{array}\right.---\left(8\right)$

In above formula, with a p respectively _l, p _rat I ₁image and image I ₂homogeneous coordinates under coordinate system, for the homogeneous coordinates under required world coordinate system, for projection matrix the i-th row jth column element, solving equations (8), the coordinate of acquisition is the three dimensional space coordinate under world coordinate system.

3) simulate the actual flange line of welding, obtain compensation rate.

The step specifically simulating space curve acquisition welding robot compensation rate is as follows:

3.1) adopt the volume coordinate point of Matlab Curve Fitting Toolbox to the punching press plate flange line obtained and laser rays intersection point to carry out matching, obtain a space curve, be actual flange line;

3.2) the theoretical flange line of punching press plate is obtained: the real flange line of welding is by matching, and the theoretical flange line of welding is also by a matching, the flange line of car door Catia model is got nine points, measures the spatial three-dimensional position of each point.Due to car door Catia model coordinate system different from the coordinate system of actual car door, the volume coordinate of the point therefore extracted in Catia model will be transformed in the middle of actual car door coordinate system, carries out matching afterwards obtain theoretical flange line to these spatial point;

3.3) actual flange line and theoretical flange line is rotated, calculate welding compensation rate: when parallel with the Z axis of world coordinate system respectively, obtain projection on an x-y plane, by calculating the compensation rate obtained in the x-direction and the z-direction, in like manner the compensation rate in X-direction and Z-direction and the compensation rate in Z-direction and Y-direction can be obtained respectively.

Introduce the Principle of Rotating of flange line below, by the three dimensional space coordinate on flange line, obtain the direction vector l (a of slot line, b, c), utilize interspace analytic geometry knowledge ask l and X-axis, Y-axis, Z axis angle be respectively α, β, γ, as shown in Equation 9.

$\left\{\begin{array}{c}\cos \mathrm{\&alpha;}=\frac{a}{\sqrt{a^2+b^2+c^2}}\\ \cos \mathrm{\&beta;}=\frac{b}{\sqrt{a^2+b^2+c^2}}\\ \cos \mathrm{\&gamma;}=\frac{c}{\sqrt{a^2+b^2+c^2}}\end{array}\right.---\left(9\right)$

Consult Fig. 9, for l is rotated parallel with X-Y plane, l can be made around X-axis rotation alpha angle, also can rotate β angle around Y-axis:

The new coordinate of point then in slot line after over-rotation is such as formula shown in 10 and 11:

$\left[\begin{array}{c}{x}^{\&prime;}\\ y^{\&prime;}\\ z^{\&prime;}\end{array}\right]=\left[\begin{array}{c}x\\ y\\ z\end{array}\right]\left[\begin{array}{ccc}1& 0& 0\\ 0& \cos \mathrm{\&alpha;}& -\sin \mathrm{\&alpha;}\\ 0& \sin \mathrm{\&alpha;}& \cos \mathrm{\&alpha;}\end{array}\right]=\left[\begin{array}{c}x\\ y\\ z\end{array}\right]*R_x---\left(10\right)$

$\left[\begin{array}{c}{x}^{\&prime;}\\ y^{\&prime;}\\ z^{\&prime;}\end{array}\right]=\left[\begin{array}{c}x\\ y\\ z\end{array}\right]\left[\begin{array}{ccc}\cos \mathrm{\&beta;}& 0& -\sin \mathrm{\&beta;}\\ 0& 1& 0\\ \sin \mathrm{\&beta;}& 0& \cos \mathrm{\&beta;}\end{array}\right]=\left[\begin{array}{c}x\\ y\\ z\end{array}\right]*R_y---\left(11\right)$

3. simulate the implementation phase of welding robot compensation rate:

Consult Figure 10, after acquisition welding robot compensation rate, utilize the inverse solution of robot kinematics theoretical, ask for the actual change amount of welding robot stepper motor, moved by programming Control stepper motor, finally realize welding robot and weld in correct position.The present invention adopts cross linear guides to simulate welding robot, complete the realization of compensation rate, to try to achieve side-play amount on an x-y plane, only need to make the stepper motor of cross linear guides under the control of the controller, run certain distance, the realization of compensation rate can be completed.

Embodiment

One, the camera calibration stage

Consult Fig. 2, method (the ZHANG Z Y utilizing Zhang Zhengyou to demarcate, A Flexible New Technique forCamera Calibration [R] .Microsoft Corporation, NSR-TR-98-71,1998) demarcate, first adopt CAD Software on Drawing to demarcate masterplate, calibrating template is the gridiron pattern of 9 × 9, and its specification is 270mm × 270mm.And then two video cameras 2,5 are fixed on appropriate location and keep motionless, the angle afterwards constantly between conversion demarcation masterplate and imaging plane and orientation, two video cameras gather 16 width images to demarcation masterplate.In the 16 width images obtain shooting, the video camera gridiron pattern target image in the lower left corner carries out Feature corner extraction, and each video camera gridiron pattern target image can extract 100 angle points, and 16 width images obtain 1600 characteristic angle dot image coordinates altogether.According to Zhang Zhengyou scaling method, world coordinate system is set on an angle point of the most left upper of video camera gridiron pattern target image, Z-direction coordinate figure is taken as zero, because each gridiron pattern size careful design is 30mm, therefore gridiron pattern feature angle point can accurately be learnt at the world coordinates of X and Y-direction.Therefore after the volume coordinate obtaining gridiron pattern target image characteristic angle point coordinate and corresponding point in demarcation plane, tool box can be utilized to complete demarcation, calculate the internal and external parameter of two video cameras 2,5, through can be calculated the calibration result of two video cameras 2,5:

The inner parameter of left video camera:

Focal Length：f _c＝[812.96176,810.83238]

Principal point：cc＝[344.91814,292.31431]

Skew：alpha_c＝[0.00000]±[0.00000]

Distortion：k _c＝[0.02792-0.17838 0.00040-0.00068 0.00000]±[0.02017 0.11745 0.00208 0.00386 0.00000]

Pixel error：err＝[0.04493 0.17823]

The external parameter of left video camera:

Translation vector：Tc_ext＝[-172.12545 -113.18676]

Rotation vector：omx_ext＝[2.04384 2.244458 -0.08679]

Rotation matrix： $\mathrm{Rc}\_\mathrm{ext}=\left[\begin{array}{ccc}-0.09110& 0.99508& 0.03891\\ 0.98910& 0.09495& -0.11250\\ -0.11564& 0.02823& -0.99288\end{array}\right]$

The inner parameter of right video camera:

Focal Length：fc＝[854.19990 855.52040]

Principal point：cc＝[320.70415 204.85482]

Skew：alpha_c＝[0.00000][0.00000]＝>angle of pixel axes＝90.00000±0.00000 degrees

Distortion：kc＝[0.05267 -0.31661 -0.00501 -0.00705 0.00000]±[0.02816 0.27114 0.00200 0.00188 0.00000]

Pixel error：err＝[0.13660 0.15065]

The external parameter of right video camera:

Translation vector：Tc_ext＝[-73.42087 -33.29384 814.86943]

Rotation vector：omx_ext＝[1.99324 2.15626 -0.16396]

Rotation matrix： $\mathrm{Rc}\_\mathrm{ext}=\left[\begin{array}{ccc}-0.07074& 0.99496& 0.07128\\ 0.97274& 0.08437& -0.21598\\ -0.22090& 0.05412& -0.97379\end{array}\right]$

Two, welding robot compensation rate obtains the stage

1) Image semantic classification is carried out to the initial acquisition image of car door punching press plate flange line in camera review

After the rgb image photographed by video camera changes into gray level image, the scope of taking pictures due to video camera is greater than the target zone required for article, in order to remove unnecessary information, carrying out cutting rapid extraction target area shear gray level image.

After obtaining target area image, image enhaucament is carried out to target area, to strengthen the contrast of target area inner laser line and peripheral region, carry out logical operation afterwards, reduce integral image gray scale, improve arithmetic speed, afterwards image is carried out to the adjustment of gray scale.

In order to obtain the high result of precision comparison, adopting the method for the medium filtering improved to the image noise reduction process after grey level enhancement, afterwards in order to make operational efficiency improve, being convenient to extraction and the rim detection of laser beam, binary conversion treatment is carried out to laser rays image.

2) volume coordinate of the actual flange line of punching press plate and laser rays intersection point is obtained

Image erosion process is carried out to the image after pretreated, extract the skeleton of laser rays, what adopt is that Hough (Hough) transfer pair skeleton carries out matching acquisition equation, ask for punching press plate actual flange line and laser rays intersection point image coordinate as shown in table 1, adopt Stereo matching to a large amount of two-dimensional pixel coordinate informations obtained afterwards, carry out three-dimensional reconstruction, and then it is as shown in table 2 to obtain three dimensional space coordinate.

The image coordinate of table 1 punching press plate flange line and laser rays intersection point

Left image intersecting point coordinate	Right image intersecting point coordinate
		(151.5902，66.1021)	(144.7114，66.6472)
(147.2135，91.6526)	(139.0501，91.6533)
		(140.2258，113.5380)	(132.5562，116.6625)
(137.2138，159.1865)	(128.4501，141.0633)
		(131.5872，159.1802)	(123.4547，171.0632)
(125.9536，186.6965)	(117.2000，199.2612)
		(120.9566，213.5877)	(110.9596，229.2100)
(114.7001,241.0914)	(100.9433，292.9845)
		(109.0705，276.7321)	(95.3271，328.6215)

The three dimensional space coordinate of table 2 punching press plate flange line and laser rays intersection point

Left image coordinate	Right image coordinate	Three-dimensional world coordinate
			(151.5902，66.1021)	(144.7114，66.6472)	(263.4822，56.8798，116.0723)
(147.2135，91.6526)	(139.0501，91.6533)	(253.9866，115.0401，116.1459)
			(140.2258，113.5380)	(132.5562，116.6625)	(248.0519，176.7611，116.2044)
(137.2138，159.1865)	(128.4501，141.0633)	(240.9303，234.9212，116.2675)
			(131.5872，159.1802)	(123.4547，171.0632)	(233.8086，301.3902，116.3224)
(125.9536，186.6965)	(117.2000，199.2612)	(230.2478，350.0541，116.3877)
			(120.9566，213.5877)	(110.9596，229.2100)	(223.1261，409.4021，116.4548)
(114.7001,241.0914)	(100.9433，292.9845)	(211.2567，472.3103，116.5213)
			(109.0705，276.7321)	(95.3271，328.6215)	(197.0134，539.9659，116.5542)

3) simulate space curve and obtain welding robot compensation rate

The volume coordinate point of Matlab Curve Fitting Toolbox to the punching press plate flange line obtained and laser rays intersection point is adopted to carry out matching, obtain a space curve, be actual flange line, the flange line of the Catia model of car door extracts several points, and its volume coordinate is transformed in the middle of actual car door coordinate system, carrying out matching to these spatial point afterwards obtains actual flange line, actual flange line and theoretical flange line are rotated, project to X-Y plane respectively, in X-Z plane and Z-Y plane, obtain compensation rate.

By the three dimensional space coordinate that flange line is put, the direction vector l of actual flange line and theoretical flange line can be obtained respectively ₁(10.6825 ,-115.1335 ,-0.1783), l ₂(16.6172 ,-150.7418 ,-0.1783), through calculating, actual flange line and theoretical flange line are α with the angle of X-axis, Y-axis and Z axis respectively ₁=84.698, β ₁=174.612, γ ₁=90.088, α ₂=83.709, β ₂=173.709, γ ₂=90.061.The rotating vector of actual flange line and theoretical flange wire-wound X-axis is respectively R _x1and R _x2, around the rotating vector R respectively that Y-axis rotates _y1and R _y2, four rotating vectors are as shown in (12) and (13) formula.

$R_{x1}=\left[\begin{array}{ccc}1& 0& 0\\ 0& 0.0924& -0.9957\\ 0& 0.9957& 0.0924\end{array}\right]R_{y1}=\left[\begin{array}{ccc}-0.9957& 0& -0.0926\\ 0& 1& 0\\ 0.0926& 0& 0.9957\end{array}\right]---\left(12\right)$

$R_{x2}=\left[\begin{array}{ccc}1& 0& 0\\ 0& 0.1095& -0.9939\\ 0& 0.9939& 0.1095\end{array}\right]R_{y2}=\left[\begin{array}{ccc}-0.9939& 0& -0.1103\\ 0& 1& 0\\ 0.1103& 0& -0.9939\end{array}\right]---\left(13\right)$

According to the equation of actual flange line and theoretical flange line, calculate the distance between two lines, be compensation rate, the mean distance of calculating is: d=3.2406mm, the mean distance of movement is 3.02mm in the X direction, and the mean distance of movement is in the Y direction 1.17mm.Recording solder joint to the distance of flange line through three coordinate measuring machine is 7.39mm, and theoretical solder joint is 4.32mm to the distance of theoretical flange line, so departing from desirable bond pad locations is 3.07mm, the distance offset in the X direction is 2.86mm, side-play amount is in the Y-axis direction 1.11mm, and result is as shown in table 3.

Table 3 side-play amount measurement result

3. simulate the implementation phase of welding robot compensation rate:

According to parameter and the end effector of transformation matrix and each rod member of mechanical arm, welding robot Inverse Kinematics Solution can be carried out, i.e. the rotational angle in each joint.

The present invention adopts cross linear guides to simulate welding robot, complete the realization of compensation rate, tried to achieve side-play amount on an x-y plane above, only need to make the stepper motor of cross linear guides under the control of the controller, run certain distance, the realization of compensation rate can be completed.

On linear cross guide rail, the model of supporting stepper motor is: 42HD2401, each pulse, can make stepper motor motion rotation 1.8 degree, radius of gyration r=4mm, and the rotation radian of stepper motor is L, and rotational angle is φ ₁, and the diameter of threaded rod is d=8mm, tilt angle theta=200 of screw thread.Therefore want to run L in the Y direction _y=1.17mm, then by L _y=Lsin θ, L=r φ ₁, φ ₁=0.8552 °

In like manner at X-direction stepper motor rotational angle φ ₂=2.2075 °.

Claims (6)

1. the car door spot welding robot course corrections method based on stereoscopic vision, it is characterized in that, the method uses a kind of automobile door welding compensation rate based on stereoscopic vision to obtain system, comprise be arranged on door skin streamline track oblique upper a word laser instrument, be symmetricly set on word laser instrument both sides and be positioned at two video cameras directly over door skin streamline track, and the terminal device be connected with a word laser instrument and two video cameras;

This car door spot welding robot course corrections method comprises the following steps:

Step one, camera calibration: put up described based on after the automobile door welding compensation rate acquisition system of stereoscopic vision, before actual acquisition car door surface image to be measured, first Zhang Zhengyou scaling method is utilized to demarcate two video cameras, obtain the inside and outside parameter of video camera, to set up the nonlinear relationship between actual determinand locus and image coordinate, and then calculate the volume coordinate of detected image Feature point correspondence;

Step 2, welding robot compensation rate obtain: a word laser instrument Emission Lasers tilts to beat on the actual flange line of punching press plate, laser rays generation deviation, and break, just on flange line, simulates the actual flange line of punching press plate by the point on flange line; Namely difference between car door actual flange line and theoretical flange line is equivalent to the difference between car door actual solder joint line and theoretical solder joint line;

Step 3, simulation welding robot compensation rate realize: after acquisition welding robot compensation rate, utilize the inverse solution of robot kinematics theoretical, ask for the actual change amount of welding robot stepper motor, moved by programming Control stepper motor, finally realize welding robot and weld in correct position.

2., according to a kind of car door spot welding robot course corrections method based on stereoscopic vision according to claim 1, it is characterized in that, the detailed process of described step one camera calibration is:

Utilize Zhang Zhengyou scaling method, calibrating template is the gridiron pattern of 9 × 9, and two video cameras are fixed, the angle afterwards constantly between conversion demarcation masterplate and imaging plane and orientation, and two video cameras gather 16 width images to demarcation masterplate; Feature corner extraction is carried out to camera calibration template image, each camera calibration template image can extract 100 angle points, according to Zhang Zhengyou scaling method, world coordinate system is set on an angle point of the most left upper of cross-hatch pattern picture, Z-direction coordinate figure is taken as zero, gridiron pattern feature angle point can be learnt at the world coordinates of X and Y-direction, after the volume coordinate obtaining plane picture characteristic angle point coordinate and corresponding point, calculates the internal and external parameter of video camera.

3. according to a kind of car door spot welding robot course corrections method based on stereoscopic vision according to claim 1, it is characterized in that, described step 2 welding robot compensation rate obtains and comprises following concrete steps:

1) Image semantic classification is carried out to car door punching press plate flange line image in camera review;

2) the actual flange line of punching press plate and the volume coordinate of laser rays intersection point is obtained;

3) rotate actual flange line and theoretical flange line, obtain welding compensation rate.

4., according to a kind of car door spot welding robot course corrections method based on stereoscopic vision according to claim 3, it is characterized in that, described step 1) Image semantic classification is carried out to car door punching press plate flange line image in camera review specifically comprise following process:

1.1) rgb image of camera acquisition is converted into gray level image;

1.2) obtain target area and image enhaucament is carried out to target area;

1.3) gradation of image adjustment;

1.4) image noise reduction;

1.5) binary conversion treatment of gray level image.

5., according to a kind of car door spot welding robot course corrections method based on stereoscopic vision according to claim 3, it is characterized in that, described step 2) volume coordinate of the actual flange line and laser rays intersection point that obtain punching press plate specifically comprises following process:

2.1) to through described step 1) image after process carries out Image erosion process, extracts the skeleton of laser rays;

2.2) Hough transformation obtains the image coordinate of the actual flange line of punching press plate and laser rays intersection point;

2.3) three-dimensional reconstruction is carried out to the actual flange line of pressed sheet part and laser rays intersection point.

6., according to a kind of car door spot welding robot course corrections method based on stereoscopic vision according to claim 3, it is characterized in that, described step 3) rotate actual flange line and theoretical flange line, obtain welding compensation rate and comprise following detailed process:

3.1) curve-fitting tool is adopted to described step 2) the volume coordinate point of the actual flange line of punching press plate that obtains and laser rays intersection point carries out matching, the space curve obtained and actual flange line;

3.2) the theoretical flange line of punching press plate is obtained;

3.3) rotate actual flange line and theoretical flange line, calculate welding compensation rate: actual flange line and theoretical flange line rotated, project to X-Y plane respectively, in X-Z plane and Z-Y plane, obtain the compensation rate of X-direction, Y-direction and Z-direction respectively.